Jonathan S

Ah, right, I forgot about this again. My plan was for this to get on hackage via haskell/containers#340, which is where the latest development is (including some pretty big speedups)....

> If the bias doesn't follow normal distribution, the result is not reliable. I think that's a bit overstating it: the OLS regression is an unbiased, consistent estimator of the...

The fact that you are seeing such large differences is surprising, at least to me. Criterion already uses what is essentially a subtractive model, running different numbers of iterations and...

> Am I misinterpreting the fit? Not at all for the standard timing loop as it is now. I was proposing increasing the time on every measurement to include more...

In lieu of `with_noop` functions, perhaps `iter_batched` could do something subtractive by default to reduce the per-iteration overhead, since that is where this problem becomes significant. It doesn't really need...

Fine by me.

The Rust side of the API looks good to me, though it technically isn't quite as powerful as the original C api. As for avoiding implementation details, `disasm_iter` could just...

Actually, I just remembered that copying the instruction is probably not valid, because of the detail pointer - copying the instruction would not copy its detail, so any further iterations...

Where are you seeing that documentation? I might be just reading the wrong thing or missing something, but [this](http://www.capstone-engine.org/iteration.html) says that "rather than letting the core allocate memory, user pre-allocates...

This also partially fixes #327 - it doesn't have all the tactics and instances or `mergeA` yet though.